DepTime
Open source · Python · Offline

Know what a Python dependency update breaks before you merge it.

DepTime points to each line of your code an update breaks, the release that changed it, and what to use instead. No AI model in the loop: it runs on your machine, costs nothing, and gives the same answer every time.

$ uvx deptime check-upgrade
Add the GitHub ActionView on GitHub
A terminal: git diff shows two pinned versions updated in requirements.txt, and deptime check-upgrade lists each line of the project the update breaks, with the release that changed it and what to use instead.
A real check-upgrade report from the published index.

Catch the break in review, not in production

Dependency bots tell you an update exists. DepTime tells you what it breaks in your code.

Every broken line

It finds each use of the updated packages in your code and checks it against the new release: removed names, keywords no longer accepted, newly required arguments. Each break comes with its line, the release that changed it, and what to use instead.

No AI in the loop

Each release of 53 fast-moving Python libraries is read statically, never imported or run, and compared with the one before. Same input, same answer, on your machine, for free.

On every update pull request

The GitHub Action checks Dependabot, Renovate and hand-made updates, marks each broken line on the pull request, and fails the check only when something really breaks.

Your coding agent can ask too

DepTime is an MCP server for Claude Code, Cursor, VS Code, Codex and any client that speaks MCP. Agents ask what changed in a library, in which release, and what replaces it, before writing code from memory.

$ claude mcp add deptime -- uvx deptime mcp

Rename maps

What well-known migrations renamed or removed, and what replaces each name, read from the libraries' own code.

Measured in the open

Every test set is collected before DepTime sees it, the pass bar is committed before the first run, and the misses are published, including how DepTime compares with AI models answering alone. Read the benchmark →

13 of 13pull requests that broke their project failed the check
46 of 48broken uses of a library found, each with its line
1 of 3,291uses that still worked flagged by mistake

First run on 264 real Dependabot and Renovate pull requests in open-source projects, collected before DepTime saw them: results.